TCG-VAE-based reservoir injection and production scheme optimization method, equipment and media

Abstract

The present application provides a TCG-VAE-based reservoir injection and production scheme optimization method, equipment and medium, which relates to the technical field of reservoir production optimization. The method includes: using a CMG simulator to simulate the production process of the reverse five-point water injection method and construct a data set; building a neural network for predicting future reservoir production data; using the data set to train and verify the neural network; based on the neural network for predicting future reservoir production data, fine-tuning the water injection amount after mutation through a water injection amount fine-tuning strategy to obtain the oil production prediction of the production well; according to the oil production prediction, calculating the net present value NPV of reservoir development, and obtaining the optimal water injection scheme. The influence of different porosity fields and permeability fields and different water injection amounts on reservoir development is taken into account, and the accuracy of simulation and prediction of reservoir production indicators is improved.

Description

Technical Field

[0001] The present application relates to the technical field of oil reservoir production optimization, and in particular to a method, device and medium for optimizing an oil reservoir injection and production scheme based on TCG-VAE. Background Art

[0002] As a method of oil production in oil reservoirs and one of the important means to improve oil recovery, water drive oil production mainly replenishes the formation pressure by injecting water to displace the oil phase in the reservoir. With the continuous development of oil fields, it is extremely challenging to achieve stable production of oil fields for a long time and obtain better development effects and higher economic benefits. Therefore, the optimization of reservoir injection and production schemes has become one of the most important issues in water drive oil production. However, the development adjustment in the traditional sense can no longer meet the needs of water drive oil production. Therefore, by combining artificial intelligence with reservoir development technology and driving with data, a scientific and effective water drive reservoir injection and production optimization method is proposed, which can greatly stimulate the development potential of water drive reservoirs and maximize their economic benefits in the production process.

[0003] Patent application CN202310942706.9 discloses a time window migration reinforcement learning injection and production optimization method based on historical control experience. First, the optimized number of wells and the optimized time step of the current reservoir target task time window are determined, and the optimal target model is obtained through neural network training. Based on the model, the current reservoir status information is read to obtain a complete production system to achieve real-time injection and production optimization.

[0004] Patent application CN202310652248.5 discloses a differentiated real-time injection and production optimization control method for a linkage intelligent injection and production device, which obtains the state data at time t+1; reads the next state and the economic net present value in this time step; assigns the next state to the current state until the production cycle to be optimized is completed; updates the action evaluation network parameters and the injection and production strategy network parameters; and according to the saved optimal model, links with the intelligent injection and production device to obtain the state information of the reservoir. However, this method has a low utilization rate of existing information, requires repeated optimization iterations, and cannot be adjusted in real time according to the current production plan.

[0005] Patent application CN202310672316.4 discloses a method and system for designing an injection and production plan for oil and water wells. Under the constraints of the total amount of water injection and the maximum water injection amount of a single well, the relationship between the water injection amount of a water well and the oil correlation is used to obtain the water injection plan with the maximum total oil production of the oil well or the maximum ratio of the total oil production of the oil well to the total water production of the oil well. This plan is the optimal plan. However, this method requires few types of data, and when the amount of single data is small, the accuracy of the predicted data is low.

[0006] Patent application CN202310465041.7 discloses a multi-agent injection-production optimization method based on adaptive basis function selection. The net present value of the initial sample is calculated using a reservoir numerical simulator to construct a database. The radial basis function network agent model is constructed using multiple basis functions using the samples in the database and the root mean square error is calculated. Multi-agent optimization is performed based on a non-dominated sorting genetic algorithm, and the optimal development plan and net present value are output, which improves the efficiency of injection-production optimization and formulates the optimal reservoir development plan.

[0007] In 2023, Zhu Bilei proposed a reservoir development index prediction model based on CNN-Bi LSTM, which built oil production and water content prediction models for single wells in the block, optimized the water injection volume of key wells using MSSSA, optimized the design of the injection and production allocation plan with the greatest development efficiency, and maximized the economic benefits of reservoir development. In 2021, Wu Junda established the TCNN water drive proxy model, which regards different production systems and production times as high-level features of the oil saturation and pressure distribution field in the whole area, and restores the abstract features of the image to the original image through convolution and transposed convolution, and restores the oil saturation and pressure distribution field images, thereby realizing the prediction of water drive production dynamics. In 2022, Du Leting combined the well-to-well connectivity model of adaptive graph correction with the time series neural network to establish a production dynamic prediction model based on graph correction, and predicted the oil production rate and water content of five oil reservoirs and the liquid production of Block B of Bohai A Reservoir. Compared with LSTM, GRU and T-GCN networks, the production dynamic prediction model based on graph correction takes into account the changes in reservoir conditions and the prediction results are more accurate. In 2022, Du Enda used GCN-LSTM to simultaneously extract time and space correlations to establish a single well production prediction model, and the results were better than those of the production prediction model established by using LSTM / GCN to extract only time or space correlations. In 2021, Tang Lihui proposed a production system optimization method based on particle swarm algorithm for new oil fields; for old oil fields, a multi-input deep neural network was constructed based on long short-term memory neural network to establish a production dynamic prediction model, and based on this model, a production system optimization method for old oil fields was proposed. In 2018, Wang Xiang analyzed the efficiency of various algorithms such as generalized pattern search, particle swarm optimization, adaptive covariance matrix evolution algorithm, and multi-level coordinated search algorithm in solving injection-production optimization problems, and proposed an injection-production splitting method that can be used to predict the water drive state of irregular well networks in heterogeneous oil reservoirs.

[0008] However, porosity and permeability directly affect the efficiency of reservoir development. If the water injection volume suddenly increases or decreases, the oil production will also increase or decrease to a certain extent, and the economic benefits of the reservoir will also be affected. The above methods have contributed to the optimization of reservoir injection and production plans to a certain extent, but none of them take into account the impact of different water injection volumes, different porosity fields and permeability fields on the economic benefits of reservoir development. Summary of the invention

[0009] The purpose of the present invention is to provide a method, device and medium for optimizing the injection and production scheme of oil reservoirs based on TCG-VAE in order to solve the problem that the prior art does not consider the impact of different water injection volumes, different porosity fields and permeability fields on the economic benefits of oil reservoir development.

[0010] The above-mentioned purpose of the present application is achieved through the following technical solutions:

[0011] S1: Use CMG simulator to simulate the production process of the reverse five-point water injection method and construct a data set;

[0012] S2: Building a neural network to predict future reservoir production data; using the data set to train and validate the neural network;

[0013] S3: Based on the neural network that predicts future reservoir production data, the water injection rate after the mutation is fine-tuned through the water injection rate fine-tuning strategy to obtain the oil production prediction of the production well;

[0014] S4: Based on the oil production forecast, calculate the net present value (NPV) of reservoir development and obtain the optimal water injection plan.

[0015] Optionally, step S1 includes:

[0016] S11: Generate porosity and permeability fields using the sequential Gaussian simulation algorithm in the CMG simulator;

[0017] S12: simulating the oil reservoir production process under the porosity field and permeability field until the size change of various production data indicators of the oil reservoir reaches a preset threshold;

[0018] S13: using an artificial neural network ANN to replace the model of water injection amount mutation and fine-tuning in the CMG simulator, adjusting the water injection amount of the CMG simulator, and continuing to simulate the oil reservoir production process until the size change of various production data indicators of the oil reservoir reaches a preset threshold, and obtaining each oil reservoir production model; the input of the artificial neural network ANN is the water injection amount and pressure of four water injection wells, the oil production and pressure of one production well, a two-dimensional porosity field and a two-dimensional permeability field, and the output of the artificial neural network ANN is the pressure of the four water injection wells, and the oil production and water production of one production well;

[0019] S14: Build a data set through various reservoir production models.

[0020] Optionally, step S2 includes: constructing a time series convolutional graph-variational autoencoder neural network model, that is, a neural network for predicting future oil reservoir production data, through a time series convolutional neural network, a graph convolutional neural network and a variational autoencoder.

[0021] Optionally, step S3 includes:

[0022] S31: The moment when the injection volume suddenly changes is taken as the starting point of time, i.e., t=1. The injection volume and pressure of the injection wells, the oil production and water production of the production wells, the two-dimensional porosity field and the two-dimensional permeability field at this moment and the past n-1 moments, a total of n moments, are considered as the input of the neural network for predicting future reservoir production data; the output of the neural network is the pressure of the injection wells, the oil production and water production of the production wells at the next i moments;

[0023] S32: if it is determined that the pressure of the water injection well at the next i moments exceeds the upper limit pressure of the formation, the water injection amount of the water injection well whose pressure exceeds the threshold at the tth moment is reduced according to a preset ratio, and the water injection amount after the sudden change is fine-tuned;

[0024] S33: Repeat step S31 until the pressure of all water injection wells is less than a preset pressure threshold, which is the upper limit pressure of the formation.

[0025] Optionally, step S4 includes:

[0026] S41: Starting from the sudden change of water injection volume, the NPV of the reservoir model simulated by CMG and the NPV of the reservoir production model predicted and fine-tuned by TCG-VAE are calculated respectively. The formula is as follows:

[0027]

[0028] Where: x w is the water injection volume during mutation; N t is the total time of reservoir simulation after the mutation; N o is the number of production wells; N w is the number of injection wells; N is the total number of production wells and injection wells; Δt n is the time step length of the nth time step; t n is the cumulative time at the end of the nth time step; b is the discount rate; r o is the crude oil sales price; is the oil production of production well i in the nth time step; c pw Cost of produced water treatment; is the water production of production well i in the nth time step; c iw the cost of filling injection wells with water; is the water injection volume of injection well j in the nth time step; C is the development cost of each well;

[0029] S42: Calculate the income and expenditure during the oil reservoir production process, and formulate the optimal water injection plan based on the net present value (NPV).

[0030] Optionally, the method further includes: introducing transfer learning to train parameters of a neural network for predicting future reservoir production data.

[0031] An electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes a reservoir injection and production scheme optimization method based on TCG-VAE.

[0032] A computer-readable storage medium stores instructions. When the instructions are executed, a method for optimizing oil reservoir injection and production plans based on TCG-VAE is executed.

[0033] The beneficial effects of the technical solution provided by this application are:

[0034] A large number of reservoir production models were generated using the CMG simulator, and a data set was constructed. The sample data set after time series segmentation was used to train the TCG-VAE neural network model. Then, according to the formation pressure limit, the water injection volume was fine-tuned to obtain the prediction and fine-tuning results of the reservoir production indicators, and the NPV net present value was calculated to obtain the optimal injection-production plan. This took into account the impact of different porosity fields and permeability fields and different water injection volumes on reservoir development, and achieved the simulation and prediction of reservoir production indicators in the next five years, which had a positive impact on the economic benefits of reservoir development. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present application will be further described below with reference to the accompanying drawings and embodiments, in which:

[0036] Figure 1 It is a step diagram of the method for optimizing the oil reservoir injection and production plan based on TCG-VAE in the embodiment of the present application;

[0037] Figure 2 It is a structural diagram of an inverse five-point water injection well pattern of an oil reservoir injection and production scheme optimization method based on TCG-VAE in an embodiment of the present application;

[0038] Figure 3 It is a structural diagram of a TCG-VAE neural network model of a reservoir injection and production scheme optimization method based on TCG-VAE in an embodiment of the present application;

[0039] Figure 4 It is a training process diagram of the TCG-VAE neural network model of the reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0040] Figure 5 It is a violin diagram of the test performance error of the TCG-VAE neural network model of the reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0041] Figure 6It is a flow chart of fine-tuning water injection volume of the oil reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0042] Figure 7 It is a diagram showing the fine-tuning result of water injection volume, the pressure prediction of water injection wells and the oil production prediction result of production wells of the oil reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0043] Figure 8 It is a diagram showing the calculation results of the oil reservoir NPV net present value of the oil reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0044] Fig. 9 It is a cumulative diagram of reservoir income and expenditure of the reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0045] Fig.10 It is a cumulative diagram of the oil reservoir NPV net present value of the oil reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0046] Fig.11 It is a structural diagram of a five-point area water injection system of a reservoir injection and production scheme optimization method based on TCG-VAE in an embodiment of the present application;

[0047] Fig.12 It is a training process diagram of transfer learning of the TCG-VAE neural network model of the oil reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0048] Fig.13 It is a diagram showing the fine-tuning result of water injection volume, the pressure prediction of water injection wells and the oil production prediction result of production wells in the case of a five-point area water injection system in the oil reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application;

[0049] Fig.14 It is a schematic diagram of the electronic device structure of the reservoir injection and production scheme optimization method based on TCG-VAE in the embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.

[0051] The embodiments of the present application provide a method for optimizing oil reservoir injection and production plans based on TCG-VAE.

[0052] Please refer to Figure 1 , Figure 1 It is a step diagram of a method for optimizing an oil reservoir injection and production plan based on TCG-VAE in an embodiment of the present application, including:

[0053] S1: Use CMG simulator to simulate the production process of the reverse five-point water injection method and construct a data set;

[0054] Step S1 includes:

[0055] S11: Generate porosity and permeability fields using the sequential Gaussian simulation algorithm in the CMG simulator;

[0056] S12: simulating the oil reservoir production process under the porosity field and permeability field until the size change of various production data indicators of the oil reservoir reaches a preset threshold;

[0057] S13: using an artificial neural network ANN to replace the model of water injection amount mutation and fine-tuning in the CMG simulator, adjusting the water injection amount of the CMG simulator, and continuing to simulate the oil reservoir production process until the size change of various production data indicators of the oil reservoir reaches a preset threshold, and obtaining each oil reservoir production model; the input of the artificial neural network ANN is the water injection amount and pressure of four water injection wells, the oil production and pressure of one production well, a two-dimensional porosity field and a two-dimensional permeability field, and the output of the artificial neural network ANN is the pressure of the four water injection wells, and the oil production and water production of one production well;

[0058] S14: Build a data set through various reservoir production models.

[0059] Specifically, the sequential Gaussian simulation algorithm was used in the CMG simulator to generate 200 porosity and permeability fields with a size of 75 × 75 (e.g. Figure 2 As shown in Figure 2, the reservoir production process for 10 years was simulated under different porosity and permeability fields. By the 10th year, the various production data indicators of the reservoir (such as the pressure and injection volume of the injection well, the oil production and water production of the production well) tended to be stable. When the simulation time reached the 11th year, the injection volume of the injection well was adjusted to 25, 30, 35, 40, 45, 50, 55, and 60m 3 / day. Considering that the injection volume of water injection wells under constant pressure production conditions is reduced by formation pressure, the CMG simulator is used to simulate the reservoir production process for 5 years. The various production data of the reservoir will stabilize again, and finally 1600 reservoir production models are obtained. Compared with the CMG simulator, the artificial neural network (ANN) has a faster running speed, so it is very necessary to use ANN to replace the model of water injection volume mutation and fine-tuning in the CMG simulator. Finally, the data set is divided into training set and test set according to 8:2.

[0060] S2: Building a neural network to predict future reservoir production data; using the data set to train and validate the neural network;

[0061] Step S2 includes: constructing a time series convolutional graph-variational autoencoder neural network model, that is, a neural network for predicting future reservoir production data, through a time series convolutional neural network, a graph convolutional neural network and a variational autoencoder.

[0062] Specifically, for the study of reservoir injection and production optimization problems, it is necessary to use existing production data to predict future production data. This method is called time series analysis. Use the data of the past 20 time segments to predict the data of the next 2 time segments, where one time segment is 30 days. The sliding window transformation is used to divide the time series data according to a fixed window size, and then the window slides on the time series data with a certain step size to obtain a series of supervised learning samples. In this application, the sliding window size is 22 (past 20 sequences + future 2 sequences), and the sliding step size is 1. Compared with the above time series, the two-dimensional porosity field and the two-dimensional permeability field are heterogeneous data and cannot be directly used for time series analysis. The variational autoencoder (VAE) is used to extract the latent vectors in the two-dimensional heterogeneous data, and the latent vectors are fused with the data after time series analysis to achieve the fusion of multi-source heterogeneous data.

[0063] Specifically, considering the connectivity between injection wells and production wells, the well network structure of the inverse five-point method can be regarded as a graph structure, where each well is a node. Assuming that there is no connectivity between injection wells and production wells cannot point to injection wells, then the injection well pointing to the production well is an edge. In order to simultaneously capture the dependencies between temporal data, spatial data, and two-dimensional heterogeneous data, a temporal convolutional graph-variational autoencoder neural network model (TCG-VAE) is proposed (e.g. Figure 3 As shown in Figure 1). Graph convolutional neural networks (GCN) make up for the defects of traditional deep learning methods in processing non-Euclidean structure data. Both time series convolutional neural networks (TCN) and recurrent neural networks (RNNs) can process time series data. Compared with RNNs, TCN effectively avoids the problems of gradient explosion and gradient disappearance. TCN is used to capture the temporal features of spatiotemporal graph data, and the output of TCN is used as the input of GCN. GCN is used to extract the spatial information of the graph, and residual connections are introduced before TCN processes the data and after GCN processes the data. The method of combining TCN and GCN to process spatiotemporal graph data is called TCGCN. After multiple TCGCN layers, the spatiotemporal features of the spatiotemporal graph are obtained. Then the latent vector extracted by VAE is fused with the spatiotemporal features, and the fused data is passed through a linear layer and a LeakyReLU activation function. Finally, the attention mechanism is introduced to weight the data of different channels to improve the output effect of the model.

[0064] Specifically, TCGCN and VAE are jointly trained, and the loss function of the temporal convolutional graph-variational autoencoder neural network model (TCG-VAE) is the mean square error (MSE) between the model output and the data generated by the CMG simulator plus the loss function of VAE. VAE It is divided into reconstruction loss and KL divergence regularization term, that is,

[0065]

[0066] Where MSE(X,X′) represents the reconstruction loss, which is the mean square error between the vector obtained by decoding the two-dimensional porosity field and permeability field and the input vector; Represents the KL divergence regularization term, which makes the latent vector generated by the encoder conform to the standard normal distribution as much as possible.

[0067] The Adam optimizer is used to update the network parameters of TCG-VAE. The network parameters with the minimum loss on the validation set are used for subsequent prediction tasks. 200, 400, 600, 800, 1000, 1200, 1400, and 1600 CMG simulated reservoir models are used as inputs of TCG-VAE and trained under the same parameter settings (e.g. Figure 4 As shown), the optimal network parameter model is obtained.

[0068] The CMG simulator was then used to simulate one porosity field and one permeability field, eight water injection methods (25, 30, 35, 40, 45, 50, 55, 60 m 3 / day) as the test case, and calculate its error MSE, such as Figure 5 As shown, it can be obtained that 1200 is the optimal number of reservoir production models. Therefore, the optimal TCG-VAE network parameter model obtained by training 1200 reservoir models is selected for subsequent testing.

[0069] S3: Based on the neural network that predicts future reservoir production data, the water injection rate after the mutation is fine-tuned through the water injection rate fine-tuning strategy to obtain the oil production prediction of the production well;

[0070] Step S3 includes:

[0071] S31: The moment when the injection volume suddenly changes is taken as the starting point of time, i.e., t=1. The injection volume and pressure of the injection wells, the oil production and water production of the production wells, the two-dimensional porosity field and the two-dimensional permeability field at this moment and the past n-1 moments, a total of n moments, are considered as the input of the neural network for predicting future reservoir production data; the output of the neural network is the pressure of the injection wells, the oil production and water production of the production wells at the next i moments;

[0072] S32: if it is determined that the pressure of the water injection well at the next i moments exceeds the upper limit pressure of the formation, the water injection amount of the water injection well whose pressure exceeds the threshold at the tth moment is reduced according to a preset ratio, and the water injection amount after the sudden change is fine-tuned;

[0073] S33: Repeat step S31 until the pressure of all water injection wells is less than a preset pressure threshold, which is the upper limit pressure of the formation.

[0074] Specifically, during the oil reservoir production process, it is necessary to use TCG-VAE to simulate the sudden change of water injection volume at a certain moment and the process in which the water injection volume gradually stabilizes under the limitation of formation pressure.

[0075] like Figure 6 As shown in the figure, the moment when the injection volume suddenly changes is taken as the starting point of time, that is, t=1. Considering this moment and the past 19 moments, the injection volume and pressure of the injection wells, the oil production and water production of the production wells, and the two-dimensional porosity field and permeability field at a total of 20 moments are used as the input of TCG-VAE, and the pressure of the injection wells, the oil production and water production of the production wells at the next two moments are obtained. If the predicted pressure of the injection well exceeds a threshold (20MPa), that is, the upper limit pressure of the formation, then the injection volume of the injection well whose pressure exceeds the threshold at the tth moment will be reduced by a certain proportion, and the above TCG-VAE calculation and fine-tuning of the injection volume will be repeated until the pressure of all injection wells is less than the threshold. Then repeat the above steps for the data at t=t+1. After completing the operations of all time steps, the pressure prediction of the injection wells, the fine-tuning of the injection volume of the injection wells, and the oil production prediction of the production wells are obtained. As shown in the attached figure Figure 7 As shown in a, the water injection volume after fine-tuning is consistent with the data generated by the CMG simulator and gradually stabilizes; Figure 7 As shown in Figure 5, the oil production predicted by TCG-VAE can fit well with the data generated by the CMG simulator.

[0076] S4: Based on the oil production forecast, calculate the net present value (NPV) of reservoir development and obtain the optimal water injection plan.

[0077] Step S4 includes:

[0078] S41: Starting from the sudden change of water injection volume, the NPV of the reservoir model simulated by CMG and the NPV of the reservoir production model predicted and fine-tuned by TCG-VAE are calculated respectively. The formula is as follows:

[0079]

[0080] Where: x w is the water injection volume during mutation; N t is the total time of reservoir simulation after the mutation; N ois the number of production wells; N w is the number of injection wells; N is the total number of production wells and injection wells; Δt n is the time step length of the nth time step; t n is the cumulative time at the end of the nth time step; b is the discount rate; r o is the crude oil sales price; is the oil production of production well i in the nth time step; c pw Cost of produced water treatment; is the water production of production well i in the nth time step; c iw the cost of filling injection wells with water; is the water injection volume of injection well j in the nth time step; C is the development cost of each well;

[0081] S42: Calculate the income and expenditure during the oil reservoir production process, and formulate the optimal water injection plan based on the net present value (NPV).

[0082] Specifically, Figure 8 As shown in Figure 2, the prediction and fine-tuning accuracy of the TCG-VAE model reached more than 95%. The income and expenditure during the reservoir production process were calculated separately, and the sudden change in water injection volume corresponding to the earliest payback time was obtained (see Appendix Fig. 9 , Fig.10 ), as attached Fig. 9 , Fig.10 As shown in this example, when the water injection volume is 40m 3 / day, the NPV is positive and the payback time is the earliest.

[0083] The method also includes: introducing transfer learning to train parameters of a neural network that predicts future reservoir production data.

[0084] Specifically, in order to verify the generalization ability of TCG-VAE, four anti-five-point water injection well networks are combined into a five-point area water injection system, such as Fig.11 As shown in the figure, there are 9 injection wells and 4 production wells. Therefore, the five-point area injection system can be divided into four areas for TCG-VAE prediction and fine-tuning. In order to make the output of TCG-VAE more accurate, transfer learning is introduced. On the basis of loading the above optimal TCG-VAE network model, the parameters of the network model are retrained, as shown in the figure. Fig.12 As shown in the figure. After the introduction of transfer learning, TCG-VAE can draw inferences from one example, without having to prepare a large number of data samples, thus reducing the training time. Considering different porosity fields and permeability fields and different mutation water injection volumes (30, 60m 3 / day), and used CMG to simulate 8 oil reservoir production processes, of which 6 models were used for training and 2 models were used for testing. The test results show that TCG-VAE after the introduction of transfer learning is more accurate in prediction and fine-tuning than before the introduction of transfer learning, such as Fig.13 As shown in the attached Fig.13 c and attached Fig.13 As shown in Figure d, the water injection volume after fine-tuning using TCG-VAE is very close to the data generated by the CMG simulator, as shown in the attached figure. Fig.13 The oil production predicted by TCG-VAE shown in e is basically consistent with the data generated by the CMG simulator after 2036.

[0085] The present application also discloses an electronic device. Fig.14 , Fig.14 The electronic device 500 may include: at least one processor 501 , at least one network interface 504 , a user interface 503 , a memory 505 , and at least one communication bus 502 .

[0086] The communication bus 502 is used to realize the connection and communication between these components.

[0087] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.

[0088] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0089] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0090] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for optimizing reservoir injection and production scheme based on TCG-VAE, characterized in that: The method comprises the following steps: S1: Use CMG simulator to simulate the production process of the reverse five-point water injection method and construct a data set; Step S1 includes: S11: Generate porosity and permeability fields using the sequential Gaussian simulation algorithm in the CMG simulator; S12: simulating the oil reservoir production process under the porosity field and permeability field until the size change of various production data indicators of the oil reservoir reaches a preset threshold; S13: using an artificial neural network ANN to replace the model of water injection amount mutation and fine-tuning in the CMG simulator, adjusting the water injection amount of the CMG simulator, and continuing to simulate the oil reservoir production process until the size change of various production data indicators of the oil reservoir reaches a preset threshold, and obtaining each oil reservoir production model; the input of the artificial neural network ANN is the water injection amount and pressure of four water injection wells, the oil production and pressure of one production well, a two-dimensional porosity field and a two-dimensional permeability field, and the output of the artificial neural network ANN is the pressure of the four water injection wells, and the oil production and water production of one production well; S14: constructing a data set through various reservoir production models; S2: Building a neural network to predict future reservoir production data; using the data set to train and validate the neural network; S3: Based on the neural network that predicts future reservoir production data, the water injection rate after the mutation is fine-tuned through the water injection rate fine-tuning strategy to obtain the oil production prediction of the production well; Step S3 includes: S31: The moment when the injection volume suddenly changes is taken as the starting point of time, i.e., t=1. The injection volume and pressure of the injection wells, the oil production and water production of the production wells, the two-dimensional porosity field and the two-dimensional permeability field at this moment and the past n-1 moments, a total of n moments, are considered as the input of the neural network for predicting future reservoir production data; the output of the neural network is the pressure of the injection wells, the oil production and water production of the production wells at the next i moments; S32: if it is determined that the pressure of the water injection well at the next i moments exceeds the upper limit pressure of the formation, the water injection amount of the water injection well whose pressure exceeds the threshold at the tth moment is reduced according to a preset ratio, and the water injection amount after the sudden change is fine-tuned; S33: repeat step S31 until the pressure of all water injection wells is less than a preset pressure threshold, which is the upper limit pressure of the formation; S4: Based on the oil production forecast, calculate the net present value (NPV) of reservoir development and obtain the optimal water injection plan.

2. The method for optimizing oil reservoir injection and production scheme based on TCG-VAE according to claim 1, characterized in that: Step S2 includes: constructing a time series convolutional graph-variational autoencoder neural network model, that is, a neural network for predicting future reservoir production data, through a time series convolutional neural network, a graph convolutional neural network and a variational autoencoder.

3. The method for optimizing oil reservoir injection and production scheme based on TCG-VAE according to claim 1, characterized in that: Step S4 includes: S41: Starting from the sudden change of water injection volume, the NPV of the reservoir model simulated by CMG and the NPV of the reservoir production model predicted and fine-tuned by TCG-VAE are calculated respectively. The formula is as follows: Where: x w is the water injection volume during mutation; N t is the total time of reservoir simulation after the mutation; N o is the number of production wells; N w is the number of injection wells; N is the total number of production wells and injection wells; Δt n is the time step length of the nth time step; t n is the cumulative time at the end of the nth time step; b is the discount rate; r o is the crude oil sales price; is the oil production of production well i in the nth time step; c pw Cost of produced water treatment; is the water production of production well i in the nth time step; c iw the cost of filling injection wells with water; is the water injection volume of injection well j in the nth time step; C is the development cost of each well; S42: Calculate the income and expenditure during the oil reservoir production process, and formulate the optimal water injection plan based on the net present value (NPV).

4. The method for optimizing oil reservoir injection and production scheme based on TCG-VAE according to claim 1, characterized in that: The method also includes: introducing transfer learning to train parameters of a neural network for predicting future reservoir production data.

5. An electronic device, characterized in that: It includes a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method as described in any one of claims 1-4.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method steps according to any one of claims 1 to 4 are executed.

Images